eventService: Optimize DML processing performance II

[asddongmen]

What problem does this PR solve?

Issue Number: ref #5430

Delete-heavy changefeeds that configure ignore-event = ["delete"] still decoded
ignored delete rows and appended them into DMLEvent before filtering, adding
unnecessary eventService scanner CPU during historical catch-up.

What is changed and how it works?

eventScanner now checks type-only DML filters before row decode for insert/delete
events and caches the result within each transaction. Ignored deletes bypass
DecodeToChunk and DMLEvent.AppendRow. Non-ignored deletes still use the
normal decode, append, and send path.

Check List

Tests

• Unit test
• Manual test on 140M row of TTL Deleted row, scan window off:

Scenario	Master avg	Current avg	Result
Ignore delete	338.3s	229.0s	32.3% faster 🔼
No-ignore delete	410.3s	400.7s	neutral

Questions

Will it cause performance regression or break compatibility?

No compatibility break expected. No-ignore delete keeps the normal path.

Do you need to update user documentation, design documentation or monitoring documentation?

No. This is an internal eventService optimization.

Release note

Improve performance for changefeeds that ignore delete events.

Summary by CodeRabbit

• New Features

  • Enhanced DML event filtering with per-transaction caching for improved performance.
  • Added support for filtering insert, delete, and update events by type without requiring row value decoding.

• Tests

  • Added benchmark tests to measure filtering performance for DELETE events.
  • Expanded unit test coverage for DML filtering scenarios and event-type-based decisions.

[ti-chi-bot]

Skipping CI for Draft Pull Request.
If you want CI signal for your change, please convert it to an actual PR.
You can still manually trigger a test run with /test all

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: bc830c6c-94d7-4784-b9ea-774018ec68d6

📥 Commits

Reviewing files that changed from the base of the PR and between 640d3d3 and 3c18b93.

📒 Files selected for processing (5)

• pkg/eventservice/event_scanner.go
• pkg/eventservice/event_scanner_benchmark_test.go
• pkg/eventservice/event_scanner_test.go
• pkg/filter/filter.go
• pkg/filter/filter_test.go

---

📝 Walkthrough

Walkthrough

Adds ShouldIgnoreDMLByEventType to the Filter interface, implementing startTs/table/event-type ignore decisions without decoded row values. dmlProcessor gains a per-transaction RowType ignore cache and uses this method in appendRow to skip decoding and update-splitting when event type alone determines the ignore outcome. New unit tests and benchmarks validate the fast path and cache behavior.

Changes

DML Event-Type Filter Fast Path

Layer / File(s)	Summary
Filter interface and ShouldIgnoreDMLByEventType implementation pkg/filter/filter.go	Adds ShouldIgnoreDMLByEventType to the Filter interface; implements it on filter encapsulating startTs, table, and event-type checks; refactors ShouldIgnoreDML to delegate to it and return early on ignore.
Filter unit test for ShouldIgnoreDMLByEventType pkg/filter/filter_test.go	Adds TestShouldIgnoreDMLByEventType covering startTs ignore, table inclusion/exclusion, insert-event filter, and delete-value-expr filter across four table metadata instances.
Per-transaction RowType ignore cache in dmlProcessor pkg/eventservice/event_scanner.go	Adds dmlTypeFilterCache field and constant, resets cache on startTxn, integrates cache helpers into appendRow to skip decode/split for non-update and update rows, and applies separate ignore checks to split insert/delete parts.
dmlProcessor.appendRow unit tests pkg/eventservice/event_scanner_test.go	Instruments mockMounter with decodeCount, adds countingDMLTypeFilter, and extends TestDMLProcessorAppendRow with subtests verifying delete, update, and split-update ignore paths skip decode and produce correct batch output.
Benchmarks for ignore-delete fast path pkg/eventservice/event_scanner_benchmark_test.go	Adds BenchmarkDMLProcessorIgnoreDelete and BenchmarkEventScannerIgnoreDelete with fast_path/after_decode sub-benchmarks using a disableDMLTypeFastPathFilter wrapper and a benchmark event getter/iterator.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~45 minutes

Possibly related PRs

• pingcap/ticdc#5194: Adds ignore-update-only-columns feature that modifies the same ShouldIgnoreDML path in event_scanner and filter.go which this PR refactors to add the pre-decode event-type fast path.

Suggested labels

size/L, release-note-none

Suggested reviewers

• flowbehappy
• hongyunyan
• 3AceShowHand

Poem

🐇 A cache so quick, it skips the decode,
When event type alone lights the ignore road.
Delete? Update? Filter knows before the parse—
No wasted rows, no unneeded parse.
The rabbit hops past rows at lightning speed! 🚀

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 11.11% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'eventService: Optimize DML processing performance II' directly addresses the main objective of optimizing DML event processing, which aligns with the changeset's focus on improving performance for delete-heavy changefeeds by filtering before decoding.
Description check	✅ Passed	The description comprehensively covers the problem statement (Issue #5430), implementation details, test results with performance metrics, and explicitly addresses compatibility and documentation questions as required by the template.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[gemini-code-assist]

Code Review

This pull request introduces a fast-path optimization in the event scanner's DML processor to check if DML events (specifically Insert and Delete) can be ignored based on their event type before performing expensive row decoding. It caches these pre-decode filter results within each transaction and adds corresponding benchmarks and unit tests. The feedback suggests addressing the magic number 3 used for the cache array size by defining a named constant or adding explanatory comments, and explores extending this fast-path optimization to the Update path (both split and non-split updates) to further enhance performance under update-heavy workloads.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[asddongmen]

/test all

[asddongmen]

/retest

[ti-chi-bot]

@asddongmen: The following test failed, say /retest to rerun all failed tests or /retest-required to rerun all mandatory failed tests:

Test name	Commit	Details	Required	Rerun command
pull-cdc-kafka-integration-heavy	3c18b93	link	true	/test pull-cdc-kafka-integration-heavy

Full PR test history. Your PR dashboard.

Details

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. I understand the commands that are listed here.

[lidezhu]

Maybe ShouldIgnoreDMLBeforeDecode is a better name?

[asddongmen]

I prefer current name because I think filter shouldn't care if an event is decoded or not.

[lidezhu]

LGTM with minor comments

[ti-chi-bot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: lidezhu

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [lidezhu]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[ti-chi-bot]

[LGTM Timeline notifier]

Timeline:

• 2026-06-23 17:07:11.569268818 +0000 UTC m=+53715.232494331: ☑️ agreed by lidezhu.